Ssd/one queue#170144
Closed
stevendanna wants to merge 5 commits into
Closed
Conversation
Add an explicit BurstFrac field to ResourceGroupConfig and wire it through to groupInfo. Previously, the burst bucket refill fraction was derived from weight/100 in refillRMGroupBurstBuckets. Making it an explicit config field decouples burst budget from fair-share weight, which is needed for the upcoming single-queue unification where the system tenant will have a high weight but a burst fraction of 1.0 while non-system tenants will have weight=1 with a burst fraction of 0.2. Epic: none Release note: None Co-Authored-By: roachdev-claude <roachdev-claude-bot@cockroachlabs.com>
Replace the two WorkQueue refill methods (refillBurstBuckets for serverless, refillRMGroupBurstBuckets for RM) with a single refillGroupBurstBuckets(rate100, cap100) that iterates all groups and scales per-group amounts by group.burstFrac. Both strategies now recover the 100%-CPU rate from tier-0 canBurst tokens (rate100 = tokens[0][canBurst] / canBurstTarget) and forward it to the queue. serverlessStrategy gains a canBurstTarget field and applies the same rate100 to both queues. Semantic change: burst budget is now "BurstFrac of 100%-CPU rate" rather than "noBurst/4 of per-tier rate". At default settings (target=0.8, BurstFrac=0.2) they agree; at non-default they diverge, with the 100%-CPU framing being target-independent. Epic: none Release note: None Co-Authored-By: roachdev-claude <roachdev-claude-bot@cockroachlabs.com>
Define the ResourceGroupConfig that will be used for the system tenant when the CPU time token AC collapses to a single WorkQueue. The system tenant gets Weight=MaxUint32 (effectively infinite priority in fair-share ordering), BurstFrac=1.0 (full 100%-CPU burst budget), and MaxCPU=true (always qualifies for burst regardless of bucket fullness). The constant is not yet installed in the default holder seed; it will be activated in a follow-up commit that routes all work to a single queue. Co-Authored-By: roachdev-claude <roachdev-claude-bot@cockroachlabs.com>
Remove the modeStrategy interface, serverlessStrategy, and rmStrategy types. Replace with direct methods on cpuTimeTokenAllocator: - computeTargets(noBurstFracs, burstDelta): pure transform from values - readTargetSettings(): reads KVCPUTimeUtilTarget and mirrors to both tiers - refillBurst(): loops over all queues unconditionally - setMode(mode): sets currentMode only Both modes now use a single utilization target setting. The old app_tenant setting (key admission.cpu_time_tokens.target_util.app_tenant, default 0.80) is retained as the registration key and given the canonical name admission.cpu_time_tokens.target_util via WithName. Existing serverless clusters that have tuned the app_tenant setting keep their value. The previously unused KVCPUTimeUtilTarget (key admission.cpu_time_tokens.target_util, default 0.75) and the per-tier KVCPUTimeSystemUtilGoal are removed. System tenant priority now comes entirely from weight and MaxCPU config, not from a higher per-tier utilization target. Co-Authored-By: roachdev-claude <roachdev-claude-bot@cockroachlabs.com>
Route all CPU time token AC work through a single WorkQueue. Remove the two-tier resource architecture that previously gave the system tenant a separate queue with higher utilization targets. Structural changes: - Remove resourceTier type, systemTenant/appTenant constants, and numResourceTiers. The granter has [numBurstQualifications] buckets (canBurst, noBurst) instead of [numResourceTiers][numBurstQualifications]. - Remove cpuTimeTokenChildGranter; cpuTimeTokenGranter implements the granter interface directly. - Collapse rates, tokenCounts, capacities, minimums, targetUtilizations from 2D to 1D. - Collapse per-tenant metrics from per-tier arrays to single counters. Per-bucket metrics go from 4 to 2 counters. Routing changes: - GetKVWorkQueue/GetCTTWorkQueue always return the single queue. - Remove useResourceGroup bool and setUseResourceGroup. Resource manager mode routing is determined by reading cpuTimeTokenACMode directly in groupKeyForWorkLocked. - Remove activeMode, currentMode, setMode, configureQueue. System tenant priority comes from systemTenantGroupConfig (Weight=MaxUint32, MaxCPU=true, BurstFrac=1.0) installed in the config holder seed, not from a separate granter tier. Co-Authored-By: roachdev-claude <roachdev-claude-bot@cockroachlabs.com>
Contributor
|
Merging to
After your PR is submitted to the merge queue, this comment will be automatically updated with its status. If the PR fails, failure details will also be posted here |
|
Your pull request contains more than 1000 changes. It is strongly encouraged to split big PRs into smaller chunks. 🦉 Hoot! I am a Blathers, a bot for CockroachDB. My owner is dev-inf. |
Member
|
This change is |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.